Bias , Confounding & Interaction

DR VIKRANT KABIRPANTHI II YEAR RESIDENT DEPARTMENT OF COMMUNITY MEDICINE NSCB MEDICAL COLLEGE JABALPUR(M.P) EMAIL:- [email protected]

Definition
Any systematic(non random)

error in the design, conduct or analysis of a study that result in a mistaken estimate of an exposures effect on the risk of disease.

Cont
Bias is a result of an error in the design or conduct of a

study. Efforts should therefore be made to reduce or eliminate bias or, at the very least, to recognize it and take it into account when interpreting the findings of a study. In bias, the focus is on an artifact of some part of the research process (assembling subjects, collecting data, analyzing data) that produces a spurious result. Bias can produce either a type 1 or a type 2 error, but we usually focus on type 1 errors due to bias.

Selection bias
The way in which cases and controls, or exposed and

non exposed individuals, were selected is such that an apparent association is observedeven if, in reality, exposure and disease are not associatedthe apparent association is the result of selection bias. In a case control study, the major source of selection bias is the manner cases or controls or both are selected and the extent to which the presence (or absence) of exposure may influence such selection

 In cohort and experimental studies, the major source

of selection bias is non - response / withdrawal from the study / losses to follow - up. In a cross - sectional study (as also in a case control study), the primary source of selection bias is selective survival, because only those who are alive can be included in such studies

The following are the ways in which selection bias can occur
1.
2. 3. 4. 5. 6. 7.

Self selection bias / Volunteers induced bias:Berksons bias (hospital selective admission) Incidence - prevalence bias (Syn - Survivorship bias, Neymans bias) Healthy worker effect Exposure related bias Bias due to loss to follow up Bias due to selection of inappropriate control group

Exclusion bias
It results from the investigators applying different

eligibility criteria to the cases and to the controls in regard to which clinical conditions in the past would permit eligibility in the study and which would serve as the basis for exclusion

 One form that selection bias can take results from non

response of potential study subjects. For example, if we are studying the possible relationship of an exposure and a disease and the response rate of potential subjects is higher in people with the disease who were exposed than in people with the disease who were not exposed, an apparent association could be observed even if in reality there is no association In general, people who do not respond in a study often differ from those who do in regard to many demographic, socioeconomic, cultural, lifestyle, and medical characteristics.

 In many studies no information is obtained from the

non responders, Non response may introduce a serious bias that may be difficult to assess. So it is important to keep non response to a minimum. In addition, any non responders should be characterized as much as possible by using whatever information is available to determine ways in which they differ from responders and to gauge the likely impact of their non response on the results of the study.

 Selection bias can result in odds ratios or relative risks

that may not be correct estimates and consequently lead to non valid inferences regarding associations of exposure and disease. Selection bias is therefore an error in selecting a study group or groups within the study and can have a major impact on the internal validity of the study and the legitimacy of the conclusion

Information Bias
Information bias can occur when the

means for obtaining information about the subjects in the study are inadequate so that as a result some of the information gathered regarding exposures and/or disease outcome is incorrect.

 Given inaccuracies in methods of data acquisition, we

may at times misclassify subjects and thereby introduce a misclassification bias.

misclassified as controls, and some without the disease (controls) may be misclassified as cases.

 This may be due to limited sensitivity and specificity of

the diagnostic tests involved or from inadequacy of information derived from medical or other records. Another possibility is that we may misclassify a persons exposure status:- we may believe the person was exposed when this was not the case, or we may believe that the person was not exposed when, in fact, exposure did occur

 Misclassification may occur in two forms:

Differential :- In differential misclassification, the rate of misclassification differs in different study groups. For example, misclassification of exposure may occur such that cases are misclassified as being exposed more often than controls are

Non differential:- Non differential misclassification results from the degree of inaccuracy that characterizes how information is obtained from any study group either cases and controls or exposed and non exposed persons. Such misclassification is not related to exposure status or to case or control status; it is just a problem inherent in the data collection methods. The usual effect of non differential misclassification is that the relative risk or odds ratio tends to be diluted, and it is shifted toward 1.0. In other words, we are less likely to detect an association even if one really exists.

Some Types and Sources of Information Bias

 If a population is monitored over a period of time,

disease ascertainment may be better in the monitored population than in the general population, and may introduce a surveillance bias, which leads to an erroneous estimate of the relative risk or odds ratio

Prevention of Bias
BLINDING - definitely in an experimental design;

even in a case control study or cohort study, the observer can be blinded. If possible, do not tell your research hypothesis to the subjects (helps preventing recall bias). In addition, if possible, try and take information about exposure from other sources, in addition to the subjects. In a follow - up study (cohort study or clinical trial), take a well defined population to avoid loss to follow up.

 Select two or more than 2 groups of controls in a case

control study (e.g. one from hospital and another healthy group); try and take different categories of diagnoses if selecting hospital controls. In an experimental design (clinical trial), ensure Random allocation, Blinding and Placebo control. the controls come from the same source population from where cases have come

Confounding
Confounding factor is defined as one

which is associated both with exposure & disease, & is distributed unequally with study & contr0l group.

Properties of confounding variable

outcome of the interest. It should not be in the direct chain or link between the exposure and outcome; its associations with exposure and outcome are indirect and independent. It exerts its effect because it is differentially distributed in the two groups

CASUAL RELATION

DUE TO CONFOUNDING

ALCOHAL

ALCOHAL

SMOKING

OESOPAGIAL CANCER

OESOPAGIAL CANCER

IMPORTANCE OF CONFOUNDING
A confounded relationship may still be a helpful guide

in screening populations even when we do not identify the specific etiologic agent involved. confounding is not an error in the study, but rather is a true phenomenon that is identified in a study and must be understood. failure to take confounding into account in interpreting the results of a study is indeed an error in the conduct of the study and can bias the conclusions of the study

Approaches to Handling Confounding

Randomization 2. Individual matching 3. Group matching 4. Restriction In the analysis of data: 1. Stratification 2. Adjustment
1.

 The singular drawback of randomization is that it can

be done only in an experimental design (e.g. drug trial, vaccine trial etc.); however, it is not applicable to most of the cause - effect research that we do in clinical practice( One group smoke & other not). The difficulty with restriction is that one tends to exclude out a lot of potential subjects, thus increasing the cost and effort of study; Secondly, the effect of the variables on which restriction has been done can not be studied

 This method in which we match one for one (i.e. for every

subject or case, we take a control who is similar to that case in respect of the confounding variable), is called as Pair Matching The second method of matching is to do a group matching or frequency matching. Suppose we want to match on 3 variables (tobacco use, age and sex) . Let us say, out of 100 cases we have 25 of them as 40 - 50 years old female tobacco users. We will then select out an equal number of controls who fit into this criteria, i.e. 25 healthy females who are 40 - 50 years old and tobacco users as controls

Adjustment During Analysis

statistical procedures and calculate the adjusted estimates, which give us the estimate of risk due to the exposure variable, after adjusting for the effect of the confounding variable If the risk in individual stratum is the same as overall risk, then there is no confounding On the other hand, if the odds ratios in the strata are very different from the overall OR ,we would conclude that there is confounding

 Limitation of stratified analysis:-if there are a large

number of confounding factors, then a large number of strata will have to be made and the individual figures in the individual strata will become very small, often zero Multiple regression analysis in such cases.

INTERACTION
When the incidence rate of disease in the presence of

two or more risk factors differs from the incidence rate expected to result from their individual effects The effect can be greater than what we would expect (positive interaction, synergism) or less than what we would expect (negative interaction, antagonism) The problem is to determine what we would expect to result from the individual effects of the exposures.

 Is there an association ?
If so, it is due to confounding ? Is there an association equally strong in strata formed

on the basis of a third party variable ?

interaction present

NO

YES

interaction absent

There are two type of models proposed in intraction

2) Multiplicative model.(Relative Risk model)

*any effect greater than additive as evidence of positive interaction, which is also called synergism

Conclusion
Biases reflect inadequacies in the design or conduct of

a study and clearly affect the validity of the findings. Biases therefore need to be assessed and, if possible, eliminated

Confounding and interaction, on the other hand,

describe the reality of the interrelationships between certain factors and a certain outcome

 Confounding and interaction characterize virtually

every situation in which etiology is addressed, because most causal questions involve the relationships of multiple exposures and multiple, possibly etiologic, factors. Such relationships are particularly important in investigating the roles of genetic and environmental factors in disease causation and in assigning responsibility for adverse health outcomes from environmental exposures